### Table 12: Pseudo-OOS Backtest (1990-2010 → 2024)

**Panel A: Calibration Summary**

| Metric | Value |
|:--|--:|
| Countries | 23 |
| 80% CI coverage | 47.8% |
| 50% CI coverage | 17.4% |
| RMSE | 1.76 |
| Naive RMSE (no change) | 1.98 |
| Relative RMSE | 0.89 |
| MAE | 1.20 |
| Direction accuracy | 47.8% |
| Safe status accuracy | 91.3% |

**Panel B: Country-Level Results**

| Country | Rating 2010 | Actual 2024 | Predicted | [P10, P90] | In 80% CI | P(safe) | Still safe |
|:--|--:|--:|--:|:--|:--|--:|:--|
| DEU | 21 | 21 | 19.5 | [18.6, 20.5] | **No** | 0.98 | Yes |
| DNK | 21 | 21 | 19.7 | [18.8, 20.6] | **No** | 1.00 | Yes |
| SWE | 21 | 21 | 19.7 | [18.7, 20.6] | **No** | 0.99 | Yes |
| NLD | 21 | 21 | 19.7 | [18.8, 20.6] | **No** | 1.00 | Yes |
| CHE | 21 | 21 | 19.8 | [18.9, 20.7] | **No** | 0.99 | Yes |
| NOR | 21 | 21 | 19.8 | [18.9, 20.7] | **No** | 1.00 | Yes |
| AUS | 21 | 21 | 19.8 | [19.0, 20.8] | **No** | 1.00 | Yes |
| CAN | 21 | 21 | 19.8 | [18.9, 20.7] | **No** | 1.00 | Yes |
| LUX | 21 | 21 | 20.1 | [19.2, 21.0] | **No** | 1.00 | Yes |
| SGP | 21 | 21 | 20.2 | [19.3, 21.0] | Yes | 1.00 | Yes |
| FIN | 21 | 20 | 19.5 | [18.6, 20.4] | Yes | 0.98 | Yes |
| USA | 21 | 20 | 19.8 | [19.0, 20.7] | Yes | 1.00 | Yes |
| HKG | 21 | 20 | 19.8 | [18.8, 20.6] | Yes | 0.99 | Yes |
| AUT | 21 | 20 | 19.8 | [18.9, 20.7] | Yes | 0.99 | Yes |
| NZL | 21 | 20 | 20.0 | [19.0, 20.7] | Yes | 1.00 | Yes |
| FRA | 21 | 19 | 19.6 | [18.6, 20.5] | Yes | 0.98 | Yes |
| QAT | 19 | 19 | 19.6 | [18.7, 20.6] | Yes | 0.99 | Yes |
| KWT | 19 | 19 | 19.6 | [18.6, 20.5] | Yes | 0.98 | Yes |
| BEL | 20 | 19 | 19.7 | [18.8, 20.6] | Yes | 1.00 | Yes |
| GBR | 21 | 19 | 19.8 | [18.8, 20.7] | Yes | 1.00 | Yes |
| IRL | 21 | 19 | 20.0 | [19.2, 20.9] | **No** | 1.00 | Yes |
| ESP | 21 | 15 | 19.8 | [18.9, 20.7] | **No** | 0.99 | **No** |
| ITA | 20 | 14 | 19.5 | [18.5, 20.4] | **No** | 0.97 | **No** |

*Model estimated on 1990-2010 data using OADR spline specification. Projections use actual 2024 demographics with training-period macro controls. 1,000 Monte Carlo draws from coefficient VCV. Naive benchmark assumes rating unchanged from 2010.*
